Apparently it's a standard that Google is pushing to replace RC4 and already is using for HTTPS between google.com and Android.
If arc catches on I'm curious if it could support inline operations. E.g. on a 100gb+ archive can I read the tar index without decrypting the entire archive first, can I extract a single file? The ChaCha algorithm is a streaming cipher, which as I understand suggests that I cannot do operations like that.
Even worse now that I'm thinking about this, if my archive has a bit error early in the file does that mean the entire archive cannot be decrypted. Maybe for long term storage I'm better off physically securing my archives than encrypting them to avoid bit rot ruining everything.
> Maybe for long term storage I'm better off physically securing my archives than encrypting them to avoid bit rot ruining everything.
Just have more copies on more/diverse media. Encrypted backups (with authenticated encryption) have the pleasant side effect of validating the backup on restore.
If you are seriously concerned with the effects of Bit Rot on long term storage I suggest you invest in Magnetic Tapes. Which can be insured against bit rot for >300 years (Yes IBM and Oracle sells insurance for this, but only on IBM tape's/decks, they'll also have to inspect your storage facility).
With highly specialized requirements comes highly specialized solutions.
The tapes themselves could last 300 years (or, for more realistic common needs, around 30 years).
However the tape reader/recorder will probably not last that long. These things need special care (don't forget the cleaning tape every other month) and can act a little weird when the mechanical parts wear down (or even before in many cases).
The Ultrium LTO standard states that a tape recorder for version N must read/write N-1 and read N-2.
Given that there is a new version around every 3 years, it gives you around 10 years of strong guaranty on your capacity to actually recover your data.
To properly manage data on a longer period you need to migrate the data to a new media.
In fact, the problem then became mostly organizational.
Your understanding is backwards. Stream ciphers are fundamentally compatible with random-access decryption, so yes this is entirely possible.
That said, authentication of a ciphertext is, in many ways, as important as encryption. So you would need to design the archive format in such a way that individual files' contents could be authenticated and decrypted on the whole, instead of authenticating the contents of the entire archive.
ChaCha is also going to be the default for OpenSSH encryption (it's already supported by upstream, but I don't know if they've changed the default yet).
Symmetric encryption is still based on secret keys, not passwords. Symmetric cryptosystems that appear to use passwords just transform them with a key derivation function into a suitable-length encryption key for the underlying cipher.
This seems counter to the "unix philosophy". I would expect to use an archiving tool piped into an encryption tool. I'm not sure of the utility of something that combines the two.
Can anyone tell me what advantage tgz has over zip? I usually curse when I have to use it, because it lacks indexes and is pretty much only good for archival tapes, if that. I wish we'd all move to a more modern format, like zip or 7zip.
I believe, technically, you can store that data in a zip's metadata headers. It is just pretty much every zip/unzip implementation doesn't support it / permission restoration breaks across OSes
Zip compresses each file separately. A folder full of zip files, each containing a single file, should take up roughly the same space as a zip archive containing all of the files. tgz doesn't do this, so it has a higher compression ratio for large collections of files. However, you cannot extract a single file alone from a tgz archive. This is why zip was designed the way it was.
zip (the version used most commonly) does not support Unicode. I have a lot of zip files created by people on Windows that turn into nonsensical filenames when extracted on Linux or OS X.
The name arc has namespace collision problems. Not in software in general (I never complained about somebody calling their language elm - that's Cantrill's job), but in archivers in particular. Arc was the format that directly preceded zip, pkzip being the program that pk started selling after it was discovered that pkarc's source code was copied verbatim from the source for the original arc utility.
Come to think of it, a variant of it is still packaged in Debian, package is called "arc".
Incidentally, for anyone choosing a name, the namecheck program from the Debian devscripts package is handy for checking if a particular name is already in use. Not perfect, but handy anyway.
Apparently it's a standard that Google is pushing to replace RC4 and already is using for HTTPS between google.com and Android.
If arc catches on I'm curious if it could support inline operations. E.g. on a 100gb+ archive can I read the tar index without decrypting the entire archive first, can I extract a single file? The ChaCha algorithm is a streaming cipher, which as I understand suggests that I cannot do operations like that.
Even worse now that I'm thinking about this, if my archive has a bit error early in the file does that mean the entire archive cannot be decrypted. Maybe for long term storage I'm better off physically securing my archives than encrypting them to avoid bit rot ruining everything.